Programmable content workflows let you describe an entire creative process as data, the shots, the gates, the budgets, the order of operations, and then run that description repeatably instead of clicking through an editor every time. In CoreReflex, that description is JSON, and it executes against a programmable engine that sits directly over editor state. Write the recipe once; run it on demand, in batch, or on a schedule, and get the same kind of output every time.
What a programmable content workflow is
A programmable content workflow is a structured, machine-readable definition of a creative job: what to make, how to judge it, and how much to spend. Where a traditional editor stores a project as opaque state you can only change by hand, a programmable workflow exposes that state as JSON you can read, diff, version, and execute. The creative becomes code, reviewable in a pull request, generated by another program, or triggered by an event.
This matters because creative work is increasingly volume work. One hero video is a craft project; two hundred localized variants, a weekly highlight reel, and a catalog of product shorts are an operations problem. Operations problems want recipes, not rituals. If you have followed our creative automation coverage, this is the layer underneath all of it.
The JSON engine over editor state
CoreReflex's editor is not a black box. Underneath the timeline, canvas, and shot list is a JSON representation of the project, and the platform ships a programmable engine that queries and commands that representation through structured calls. Anything you can do by hand, board a shot, set a camera move, attach a voiceover, change a brand color, has a structured equivalent the engine can execute and return a structured result for.
Because the engine speaks the editor's own language, a workflow you script is not a fragile macro recorded on top of a UI. It manipulates the same source of truth the app does. That is what makes a recipe portable: the JSON you author is the JSON the studio runs. For a plain-language tour of what these recipes can express, see our explainer on JSON workflow recipes in CoreReflex.
What you can define in a recipe
A workflow recipe typically declares three things, the creative, the gates, and the governance:
- Shots and assets. Each shot carries a
role(hook, b-roll, proof, CTA), acameramove, and aprompt. The same recipe can request images, music, voiceover, and video, all generated on Google Vertex AI. Veo and Kling for video, Imagen for stills, Lyria for music, plus STT and TTS for voice. - Quality gates. You attach the checks each shot must pass, prompt match, sharpness, motion coherence, on-screen text legibility, on-brand fit, claims risk, and the engine scores against them, regenerating only the shots that fail. The deeper mechanics live in our piece on quality gates in automated pipelines.
- Budgets and checkpoints. A recipe can cap how many credits a run may spend and where a human must approve before it proceeds. Planning and scoring are free; only generation costs credits, so the budget governs exactly the expensive step.
Because it is all data, you can template a recipe and vary the inputs: one base workflow, fifty product names, fifty finished shorts.
A recipe is reproducible by construction
Every generated asset carries a portable trace, model, prompt, parameters, and score. When a recipe runs, that trace is captured alongside the JSON, so a run is not a one-off you can never recreate. You can replay it, audit it, or hand it to a client. Our guide to provenance and audit trails for AI content covers why that reproducibility is the difference between a demo and a deliverable.
How to build your first programmable workflow
- Prototype in the editor. Board the film by hand once, shots, camera moves, voice, brand kit. You are designing the shape the recipe will encode.
- Export the structure. Capture the project as JSON. This is your recipe's starting point: the literal editor state expressed as data.
- Parameterize the variables. Replace the values that should change per run, product name, headline, palette, target length, with inputs. Now one recipe produces a family of outputs.
- Attach gates and a budget. Declare the quality checks each shot must clear and a credit ceiling for the run, plus any human checkpoints. The pipeline will not blow past either.
- Run it. Execute on demand to validate, then graduate to batch for volume or to a schedule for hands-off production. Scheduling turns a recipe into a standing process, see scheduling hands-free video generation.
The full surface, engine calls, recipe schema, and limits, is documented for developers in the platform docs.
On-demand, batch, and scheduled execution
The same recipe runs three ways, and the mode is a property of how you trigger it, not a different system:
| Mode | When it runs | Best for |
|---|---|---|
| On-demand | You invoke it now | Validating a recipe, one-off requests |
| Batch | Many inputs at once | Localizations, product catalogs, variant testing |
| Scheduled | On a cron-like cadence | Weekly reels, recurring social, always-on pipelines |
Batch and scheduled runs are where programmable workflows earn their keep. A producer who once assembled a weekly recap by hand instead writes the recipe once and lets Autopilot run it every Friday. For the high-volume case specifically, our practical guide to batch video generation walks through throughput, polling, and asset retrieval.
Keeping a human in the loop
Programmable does not mean unsupervised. A recipe can declare checkpoints where a person reviews and approves before the run continues, after planning, before an expensive batch, or before a client-facing render. The engine pauses, surfaces the work, and waits. That keeps autonomy and accountability in the same workflow; our piece on human-in-the-loop checkpoints in Autopilot shows how to place them where they matter.
Frequently asked questions
What are programmable content workflows?
They are creative processes defined as structured data, typically JSON, that a system can execute repeatably. Instead of manually producing each video or image, you author a recipe that declares the shots, the quality gates, and the budget, then run it on demand, in batch, or on a schedule. The output is consistent because the definition is consistent.
Can I define creative as JSON?
Yes. CoreReflex represents editor state as JSON and exposes a programmable engine over it, so a film, image, or slide deck can be authored, queried, and commanded as data. You can export a project you built by hand, parameterize the parts that should vary, and replay it as a recipe across many inputs.
What is the JSON engine over editor state?
It is the programmable layer that reads and writes the same JSON the editor uses as its source of truth. Because it speaks the editor's native structure rather than scripting the UI, the recipes you author are durable and portable, the JSON you write is the JSON the studio runs, and every action returns a structured result you can inspect.
How is this different from a no-code automation tool?
General automation tools connect apps and move data between them. A programmable content workflow goes inside the creative itself. It boards shots, scores them against quality gates, regenerates failures, and renders deterministically, all on an owned stack. The recipe governs production, not just plumbing.
Make your creative repeatable
When creative is data, it stops being a series of one-off efforts and becomes a process you can version, test, and scale. Build a recipe, attach your gates and a budget, and let it run, on demand for a single cut, in batch for a catalog, or on a schedule for an always-on pipeline. Start free with no credit card and turn your first project into a programmable workflow.